UN Report Warns AI‑Driven Data Centres to Double Power and Water Use by 2030, Raising Questions of Global Accountability

In a comprehensive assessment issued by the United Nations' Intergovernmental Panel on Digital Infrastructure, researchers have projected that the aggregate consumption of electrical power by artificial‑intelligence‑driven data centres is destined to double by the close of the decade, thereby casting a foreboding silhouette upon the lofty emissions‑reduction commitments previously articulated by the Parties to the Paris Agreement. The lead author, Professor Kaveh Madani, cautioned that the popular public discourse which confines artificial intelligence to the realm of intangible software overlooks the substantial physical infrastructure comprising vast server farms, cooling apparatus, high‑capacity transmission grids, semiconductor fabrication facilities, and the mineral extraction operations whose water footprints are now inexorably intertwined with the algorithmic surge.

According to the report's calibrated modelling, the present annual electricity demand attributable to AI‑related processing workloads approximates 300 terawatt‑hours, a figure projected to ascend to roughly 600 terawatt‑hours by 2030, while concomitant water usage for cooling and humidification is forecast to rise from an estimated 12 cubic kilometres to in excess of 24 cubic kilometres within the same interval. Such an escalation, the authors contend, surpasses the incremental growth rates historically observed in conventional cloud services by a factor of two, thereby amplifying the urgency for policy makers to reassess the assumed decoupling of digital acceleration from the physicochemical constraints of energy generation and freshwater availability.

The impending surge in power draw threatens to erode the modest gains recorded under the United Nations Framework Convention on Climate Change, as the additional carbon dioxide emissions engendered by fossil‑fuel‑based electricity generation could offset the net reductions achieved through efficiency measures in other sectors. Moreover, the amplified water withdrawal from thermoelectric cooling cycles risks exacerbating hydro‑ecological stress in river basins already strained by agricultural demand and climate‑induced variability, a circumstance that could invite diplomatic friction between riparian states sharing transboundary watercourses.

For a nation such as India, where freshwater scarcity already imposes severe constraints on industrial expansion and where the burgeoning digital economy aspires to host a significant share of global AI compute, the projected doubling of data‑centre water consumption portends a potential clash between technological ambition and the tenets of sustainable resource stewardship. Consequently, Indian regulatory bodies, including the Ministry of Electronics and Information Technology and the Central Water Commission, may be compelled to devise nuanced licensing regimes that integrate quantifiable water‑efficiency benchmarks alongside the usual energy‑intensity standards applied to new data‑centre projects.

The revelation arrives amidst ongoing multilateral deliberations at the G20 summit, where member states have expressed divergent positions on the governance of artificial intelligence, with some advocating for a light‑touch regulatory approach to preserve competitive advantage, while others, notably members of the European Union, have urged the incorporation of environmental impact assessments into any future AI‑centric trade agreements. In parallel, the United Nations' Sustainable Development Goals framework, particularly Goal 7 on affordable and clean energy and Goal 6 on clean water and sanitation, now confronts an inadvertent loophole whereby digital transformation, if left unchecked, may subvert the very objectives it was intended to advance.

If the United Nations' own estimations indicate that artificial‑intelligence‑driven data‑centre operations will double their demand for electricity and freshwater within a decade, does this not compel the signatories of the Paris Agreement to reconsider the adequacy of their nationally determined contributions in light of a newly quantified source of emissions that was previously omitted from their inventories? Furthermore, should the principle of common but differentiated responsibilities be extended to encompass the obligate sharing of mitigation costs for digital‑infrastructure water extraction, thereby obligating affluent technology exporters to finance water‑conservation programmes in downstream nations that would otherwise bear the brunt of increased abstraction? In what manner, if any, can national data‑protection statutes be harmonised with environmental law to create a coherent framework that obliges corporations to disclose the precise volumetric water footprints associated with each trained model, thereby granting civil society the evidentiary basis to demand remedial action where thresholds are exceeded?

Might the emerging disparity between the declared intent to regulate AI through soft‑law instruments such as the OECD Principles on AI and the hard‑law necessity for enforceable environmental safeguards compel nations to negotiate a binding treaty that explicitly links algorithmic compute intensity to quantifiable caps on power and water usage, and if so, which international body would possess the requisite jurisdiction to monitor compliance and impose sanctions? Lastly, could the failure of existing multilateral forums to incorporate transparent reporting mechanisms for AI‑related resource consumption be interpreted as a breach of the duty of cooperation enshrined in the UN Charter, thereby granting affected states a legal basis to seek redress before international tribunals for the alleged negligence of the global community in averting foreseeable environmental harm? Is it not incumbent upon the International Telecommunication Union, historically charged with spectrum allocation, to assume a pioneering role in instituting standardized metrics for energy and water efficiency that could be incorporated into future revisions of the World Radiocommunication Conferences' resolutions, thus ensuring that the governance of the invisible currents of data does not eclipse the visible scarcity of planetary resources?

Published: June 4, 2026